Is it true that 4.0.0 Broke all my publication quality figures?

[nickharrigan]

All the figures that I was using for publications were working three weeks ago, but now look terrible.

I spent about 3 hours trying to fix them but nothing would work.

And then ChatGPT told me it was because of the update to ggplot2 4.0.0 on 16 Sept.

Is this right? Were the changes so big that they break most people's publication quality bespoke figures?

This is going to take me quite a long time to fix, if this is the issue.

Was there the need to force the roll out relatively silently? And it seems quite hard to roll back?

Or am I just incompetent?

[thomasp85]

I wouldn't say that we have done this silently as we have written and posted quite a bit about the update. However, we should not break people's plots unless they have used some unsupported functionality. Can you share a plot that seems broken with the new version.

Also, have you confirmed that downgrading to ggplot2 3.5 solves the issue?

[nickharrigan]

I tried to downgrade but it threw a lot of errors because of dependencies of other libraries that had integrated with 4.0.0, the need to download rtools 45 to deal with these dependent libraries, and so forth. I had to give up for today. I will work through the layers of issues tomorrow and try to unpack what has gone wrong and update here. Honestly, I needed to make graphs for a collaborator and spent most of the day trying to work out why ggplots that worked a few weeks ago were unusable, ugly, and unfixable. Maybe I made a mistake, but as I said, ChatGPT said it was the silent update to 4.0.0 on about 16 Sept that caused the meltdown of my custom graphs. It could be wrong, but I will update with what I find

[thomasp85]

Not saying that it isn't because of the upgrade but ChatGPT is pretty clueless about a lot of stuff :-)

If you share some reproducible code we can help you look into it

[nickharrigan]

I've downgraded to 3.4.4

The graph below was failing and ugly under 4.0.0 but works again with 3.4.4

I've made a reproducable example:

---

library(ggplot2)

Create dataframe

df <- data.frame(
group = factor(c("0", "1-3", "4-6", "7-9", "10+"),
levels = c("0", "1-3", "4-6", "7-9", "10+")),
estimate = c(0, 0.28, 0.38, 0.50, 0.64),
lwr = c(0, 0.1627, 0.3012, 0.3792, 0.3678),
upr = c(0, 0.3973, 0.4588, 0.6208, 0.9122)
)

Graph function

bar_whisker_graph <- function(data, my.title, my.y.label, my.x.label, my_note, y.intercept, items = 5) {

if(items == 5) {
colour_variable <- 'scale_color_manual(values = c("#85ceecff", "#ffa27bff", "#ffa27bff", "#ffa27bff", "#ffa27bff"))'
} else if(items == 6) {
colour_variable <- 'scale_color_manual(values = c("#85ceecff", "#ffa27bff", "#ffa27bff", "#ffa27bff", "#ffa27bff", "#ffa27bff"))'
}

p <- ggplot(data, aes(x = group, y = estimate, color = group)) +
geom_point(size = 4, shape = 20, stroke = 1.5) +
geom_linerange(aes(ymin = lwr, ymax = upr), linewidth = 0.8) +
geom_text(aes(label = round(estimate, 2)),
vjust = 0.5, hjust = -0.5, size = 5, color = "black") +
geom_hline(yintercept = 0, linetype = "dashed", color = "black", linewidth = 0.5) +
eval(parse(text = colour_variable)) +
labs(
title = my.title,
x = my.x.label,
y = my.y.label
) +
theme_minimal(base_size = 15) +
geom_hline(yintercept = y.intercept, linetype = "dotted", color = "black") +
theme(
plot.title = element_text(hjust = 0.5, size = 18, face = "bold", margin = margin(b = 20)),
axis.title.y = element_text(size = 14, face = "bold", margin = margin(t = 10)),
axis.text = element_text(size = 12, face = "bold"),
legend.position = "none",
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
plot.margin = unit(c(1, 1, 1, 1), "cm"),
axis.line = element_line(linewidth = 0.25, colour = "black"),
axis.line.x = element_blank(),
plot.caption = element_text(hjust = 0, size = 9, face = "italic", color = "black")
) +
labs(caption = my_note)

return(p)
}

Call the function

p1 <- bar_whisker_graph(df,
"Subsequent Adaptations Increases with Prior Adaptations",
"Subsequent Adaptions (count)\n(mean)",
"\nNumber of Prior Adaptations (count)",
"\nNote: Points represent mean Subsequent Adaptations by Prior Adaptations.\nConfidence intervals are 95%.",
0
)

print(p1)

[teunbrand]

Thank you for the reprex. Below is just a render of the reprex using ggplot2 version 4.0.0. Can you point out the failing/ugly/broken parts of the plot? It isn't immediately obvious to me.

library(ggplot2)
packageVersion("ggplot2")
#> [1] '4.0.0'

df <- data.frame(
  group = factor(c("0", "1-3", "4-6", "7-9", "10+"),
                 levels = c("0", "1-3", "4-6", "7-9", "10+")),
  estimate = c(0, 0.28, 0.38, 0.50, 0.64),
  lwr = c(0, 0.1627, 0.3012, 0.3792, 0.3678),
  upr = c(0, 0.3973, 0.4588, 0.6208, 0.9122)
)

bar_whisker_graph <- function(data, my.title, my.y.label, my.x.label, my_note, y.intercept, items = 5) {
  
  if(items == 5) {
    colour_variable <- 'scale_color_manual(values = c("#85ceecff", "#ffa27bff", "#ffa27bff", "#ffa27bff", "#ffa27bff"))'
  } else if(items == 6) {
    colour_variable <- 'scale_color_manual(values = c("#85ceecff", "#ffa27bff", "#ffa27bff", "#ffa27bff", "#ffa27bff", "#ffa27bff"))'
  }
  
  p <- ggplot(data, aes(x = group, y = estimate, color = group)) +
    geom_point(size = 4, shape = 20, stroke = 1.5) +
    geom_linerange(aes(ymin = lwr, ymax = upr), linewidth = 0.8) +
    geom_text(aes(label = round(estimate, 2)),
              vjust = 0.5, hjust = -0.5, size = 5, color = "black") +
    geom_hline(yintercept = 0, linetype = "dashed", color = "black", linewidth = 0.5) +
    eval(parse(text = colour_variable)) +
    labs(
      title = my.title,
      x = my.x.label,
      y = my.y.label
    ) +
    theme_minimal(base_size = 15) +
    geom_hline(yintercept = y.intercept, linetype = "dotted", color = "black") +
    theme(
      plot.title = element_text(hjust = 0.5, size = 18, face = "bold", margin = margin(b = 20)),
      axis.title.y = element_text(size = 14, face = "bold", margin = margin(t = 10)),
      axis.text = element_text(size = 12, face = "bold"),
      legend.position = "none",
      panel.grid.major = element_blank(),
      panel.grid.minor = element_blank(),
      plot.margin = unit(c(1, 1, 1, 1), "cm"),
      axis.line = element_line(linewidth = 0.25, colour = "black"),
      axis.line.x = element_blank(),
      plot.caption = element_text(hjust = 0, size = 9, face = "italic", color = "black")
    ) +
    labs(caption = my_note)
  
  return(p)
}

bar_whisker_graph(
  df,
  "Subsequent Adaptations Increases with Prior Adaptations",
  "Subsequent Adaptions (count)\n(mean)",
  "\nNumber of Prior Adaptations (count)",
  "\nNote: Points represent mean Subsequent Adaptations by Prior Adaptations.\nConfidence intervals are 95%.",
  0
)

^{Created on 2025-10-08 with reprex v2.1.1}

[nickharrigan]

That is roughly correct.

I don't have exact parallels created under 4.0.0

But below is the type of graphs I was getting

I will revert to 4.0.0 and upload the same image on my pc

[nickharrigan]

After another five hours of debugging I found that multiple dependencies across my code were causing unexpected failures.

For example
library(showtext)
showtext_auto()

This would completely ruin some ggplot2 figures because it was running the background. And it broke figures in both 3.4.4 and 4.0.0 versions of ggplot2.

The lesson for me is modularisation and not running unsafe functions like showtext_auto()

Thanks for your tolerance of my mistake.

[teunbrand]

Thanks for reporting back on this. So if the problem appears to be showtext rather than ggplot2, it isn't something that should be fixed here and we can close this issue.

[thomasp85]

To clarify. Showtext works purely on the graphics device level so it is very unlikely that any change in ggplot2 can have affected this. You may wish to check which graphics device is being used and whether they work correctly with showtext